Test-time adaptation (TTA) seeks to tackle potential distribution shifts between training and test data by adapting a given model w.r.t. any test sample. Although recent TTA has shown promising performance, we still face two key challenges: 1) prior methods perform backpropagation for each test sample, resulting in unbearable optimization costs to many applications; 2) while existing TTA can significantly improve the test performance on out-of-distribution data, they often suffer from severe performance degradation on in-distribution data after TTA (known as forgetting). To this end, we have proposed an Efficient Anti-Forgetting Test-Time Adaptation (EATA) method which develops an active sample selection criterion to identify reliable and non-redundant samples for test-time entropy minimization. To alleviate forgetting, EATA introduces a Fisher regularizer estimated from test samples to constrain important model parameters from drastic changes. However, in EATA, the adopted entropy loss consistently assigns higher confidence to predictions even for samples that are underlying uncertain, leading to overconfident predictions. To tackle this, we further propose EATA with Calibration (EATA-C) to separately exploit the reducible model uncertainty and the inherent data uncertainty for calibrated TTA. Specifically, we measure the model uncertainty by the divergence between predictions from the full network and its sub-networks, on which we propose a divergence loss to encourage consistent predictions instead of overconfident ones. To further recalibrate prediction confidence, we utilize the disagreement among predicted labels as an indicator of the data uncertainty, and then devise a min-max entropy regularizer to selectively increase and decrease prediction confidence for different samples. Experiments on image classification and semantic segmentation verify the effectiveness of our methods.
Visual State Space Model (VMamba) has recently emerged as a promising architecture, exhibiting remarkable performance in various computer vision tasks. However, its robustness has not yet been thoroughly studied. In this paper, we delve into the robustness of this architecture through comprehensive investigations from multiple perspectives. Firstly, we investigate its robustness to adversarial attacks, employing both whole-image and patch-specific adversarial attacks. Results demonstrate superior adversarial robustness compared to Transformer architectures while revealing scalability weaknesses. Secondly, the general robustness of VMamba is assessed against diverse scenarios, including natural adversarial examples, out-of-distribution data, and common corruptions. VMamba exhibits exceptional generalizability with out-of-distribution data but shows scalability weaknesses against natural adversarial examples and common corruptions. Additionally, we explore VMamba's gradients and back-propagation during white-box attacks, uncovering unique vulnerabilities and defensive capabilities of its novel components. Lastly, the sensitivity of VMamba to image structure variations is examined, highlighting vulnerabilities associated with the distribution of disturbance areas and spatial information, with increased susceptibility closer to the image center. Through these comprehensive studies, we contribute to a deeper understanding of VMamba's robustness, providing valuable insights for refining and advancing the capabilities of deep neural networks in computer vision applications.
The proliferation of digital images and the advancements in deep learning have paved the way for innovative solutions in various domains, especially in the field of image classification. Our project presents an in-depth study and implementation of an image classification system specifically tailored to identify and classify images of Indian cities. Drawing from an extensive dataset, our model classifies images into five major Indian cities: Ahmedabad, Delhi, Kerala, Kolkata, and Mumbai to recognize the distinct features and characteristics of each city/state. To achieve high precision and recall rates, we adopted two approaches. The first, a vanilla Convolutional Neural Network (CNN) and then we explored the power of transfer learning by leveraging the VGG16 model. The vanilla CNN achieved commendable accuracy and the VGG16 model achieved a test accuracy of 63.6%. Evaluations highlighted the strengths and potential areas of improvement, positioning our model as not only competitive but also scalable for broader applications. With an emphasis on open-source ethos, our work aims to contribute to the community, encouraging further development and diverse applications. Our findings demonstrate the potential applications in tourism, urban planning, and even real-time location identification systems, among others.
Cytology image segmentation is quite challenging due to its complex cellular structure and multiple overlapping regions. On the other hand, for supervised machine learning techniques, we need a large amount of annotated data, which is costly. In recent years, late fusion techniques have given some promising performances in the field of image classification. In this paper, we have explored a fuzzy-based late fusion techniques for cytology image segmentation. This fusion rule integrates three traditional semantic segmentation models UNet, SegNet, and PSPNet. The technique is applied on two cytology image datasets, i.e., cervical cytology(HErlev) and breast cytology(JUCYT-v1) image datasets. We have achieved maximum MeanIoU score 84.27% and 83.79% on the HErlev dataset and JUCYT-v1 dataset after the proposed late fusion technique, respectively which are better than that of the traditional fusion rules such as average probability, geometric mean, Borda Count, etc. The codes of the proposed model are available on GitHub.
Histopathological whole slide image (WSI) analysis with deep learning has become a research focus in computational pathology. The current paradigm is mainly based on multiple instance learning (MIL), in which approaches with Transformer as the backbone are well discussed. These methods convert WSI tasks into sequence tasks by representing patches as tokens in the WSI sequence. However, the feature complexity brought by high heterogeneity and the ultra-long sequences brought by gigapixel size makes Transformer-based MIL suffer from the challenges of high memory consumption, slow inference speed, and lack of performance. To this end, we propose a retentive MIL method called RetMIL, which processes WSI sequences through hierarchical feature propagation structure. At the local level, the WSI sequence is divided into multiple subsequences. Tokens of each subsequence are updated through a parallel linear retention mechanism and aggregated utilizing an attention layer. At the global level, subsequences are fused into a global sequence, then updated through a serial retention mechanism, and finally the slide-level representation is obtained through a global attention pooling. We conduct experiments on two public CAMELYON and BRACS datasets and an public-internal LUNG dataset, confirming that RetMIL not only achieves state-of-the-art performance but also significantly reduces computational overhead. Our code will be accessed shortly.
Semi-supervised learning (SSL) seeks to enhance task performance by training on both labeled and unlabeled data. Mainstream SSL image classification methods mostly optimize a loss that additively combines a supervised classification objective with a regularization term derived solely from unlabeled data. This formulation neglects the potential for interaction between labeled and unlabeled images. In this paper, we introduce InterLUDE, a new approach to enhance SSL made of two parts that each benefit from labeled-unlabeled interaction. The first part, embedding fusion, interpolates between labeled and unlabeled embeddings to improve representation learning. The second part is a new loss, grounded in the principle of consistency regularization, that aims to minimize discrepancies in the model's predictions between labeled versus unlabeled inputs. Experiments on standard closed-set SSL benchmarks and a medical SSL task with an uncurated unlabeled set show clear benefits to our approach. On the STL-10 dataset with only 40 labels, InterLUDE achieves 3.2% error rate, while the best previous method reports 14.9%.
Real-world vision models in dynamic environments face rapid shifts in domain distributions, leading to decreased recognition performance. Using unlabeled test data, continuous test-time adaptation (CTTA) directly adjusts a pre-trained source discriminative model to these changing domains. A highly effective CTTA method involves applying layer-wise adaptive learning rates for selectively adapting pre-trained layers. However, it suffers from the poor estimation of domain shift and the inaccuracies arising from the pseudo-labels. This work aims to overcome these limitations by identifying layers for adaptation via quantifying model prediction uncertainty without relying on pseudo-labels. We utilize the magnitude of gradients as a metric, calculated by backpropagating the KL divergence between the softmax output and a uniform distribution, to select layers for further adaptation. Subsequently, for the parameters exclusively belonging to these selected layers, with the remaining ones frozen, we evaluate their sensitivity to approximate the domain shift and adjust their learning rates accordingly. We conduct extensive image classification experiments on CIFAR-10C, CIFAR-100C, and ImageNet-C, demonstrating the superior efficacy of our method compared to prior approaches.
Despite the availability of large datasets for tasks like image classification and image-text alignment, labeled data for more complex recognition tasks, such as detection and segmentation, is less abundant. In particular, for instance segmentation annotations are time-consuming to produce, and the distribution of instances is often highly skewed across classes. While semi-supervised teacher-student distillation methods show promise in leveraging vast amounts of unlabeled data, they suffer from miscalibration, resulting in overconfidence in frequently represented classes and underconfidence in rarer ones. Additionally, these methods encounter difficulties in efficiently learning from a limited set of examples. We introduce a dual-strategy to enhance the teacher model's training process, substantially improving the performance on few-shot learning. Secondly, we propose a calibration correction mechanism that that enables the student model to correct the teacher's calibration errors. Using our approach, we observed marked improvements over a state-of-the-art supervised baseline performance on the LVIS dataset, with an increase of 2.8% in average precision (AP) and 10.3% gain in AP for rare classes.
This paper proposes an optimization of an existing Deep Neural Network (DNN) that improves its hardware utilization and facilitates on-device training for resource-constrained edge environments. We implement efficient parameter reduction strategies on Xception that shrink the model size without sacrificing accuracy, thus decreasing memory utilization during training. We evaluate our model in two experiments: Caltech-101 image classification and PCB defect detection and compare its performance against the original Xception and lightweight models, EfficientNetV2B1 and MobileNetV2. The results of the Caltech-101 image classification show that our model has a better test accuracy (76.21%) than Xception (75.89%), uses less memory on average (847.9MB) than Xception (874.6MB), and has faster training and inference times. The lightweight models overfit with EfficientNetV2B1 having a 30.52% test accuracy and MobileNetV2 having a 58.11% test accuracy. Both lightweight models have better memory usage than our model and Xception. On the PCB defect detection, our model has the best test accuracy (90.30%), compared to Xception (88.10%), EfficientNetV2B1 (55.25%), and MobileNetV2 (50.50%). MobileNetV2 has the least average memory usage (849.4MB), followed by our model (865.8MB), then EfficientNetV2B1 (874.8MB), and Xception has the highest (893.6MB). We further experiment with pre-trained weights and observe that memory usage decreases thereby showing the benefits of transfer learning. A Pareto analysis of the models' performance shows that our optimized model architecture satisfies accuracy and low memory utilization objectives.
For privacy and security concerns, the need to erase unwanted information from pre-trained vision models is becoming evident nowadays. In real-world scenarios, erasure requests originate at any time from both users and model owners. These requests usually form a sequence. Therefore, under such a setting, selective information is expected to be continuously removed from a pre-trained model while maintaining the rest. We define this problem as continual forgetting and identify two key challenges. (i) For unwanted knowledge, efficient and effective deleting is crucial. (ii) For remaining knowledge, the impact brought by the forgetting procedure should be minimal. To address them, we propose Group Sparse LoRA (GS-LoRA). Specifically, towards (i), we use LoRA modules to fine-tune the FFN layers in Transformer blocks for each forgetting task independently, and towards (ii), a simple group sparse regularization is adopted, enabling automatic selection of specific LoRA groups and zeroing out the others. GS-LoRA is effective, parameter-efficient, data-efficient, and easy to implement. We conduct extensive experiments on face recognition, object detection and image classification and demonstrate that GS-LoRA manages to forget specific classes with minimal impact on other classes. Codes will be released on \url{https://github.com/bjzhb666/GS-LoRA}.
Image classification with neural networks (NNs) is widely used in industrial processes, situations where the model likely encounters unknown objects during deployment, i.e., out-of-distribution (OOD) data. Worryingly, NNs tend to make confident yet incorrect predictions when confronted with OOD data. To increase the models' reliability, they should quantify the uncertainty in their own predictions, communicating when the output should (not) be trusted. Deep ensembles, composed of multiple independent NNs, have been shown to perform strongly but are computationally expensive. Recent research has proposed more efficient NN ensembles, namely the snapshot, batch, and multi-input multi-output ensemble. This study investigates the predictive and uncertainty performance of efficient NN ensembles in the context of image classification for industrial processes. It is the first to provide a comprehensive comparison and it proposes a novel Diversity Quality metric to quantify the ensembles' performance on the in-distribution and OOD sets in one single metric. The results highlight the batch ensemble as a cost-effective and competitive alternative to the deep ensemble. It matches the deep ensemble in both uncertainty and accuracy while exhibiting considerable savings in training time, test time, and memory storage.
Vision Transformer (ViT) has emerged as a prominent backbone for computer vision. For more efficient ViTs, recent works lessen the quadratic cost of the self-attention layer by pruning or fusing the redundant tokens. However, these works faced the speed-accuracy trade-off caused by the loss of information. Here, we argue that token fusion needs to consider diverse relations between tokens to minimize information loss. In this paper, we propose a Multi-criteria Token Fusion (MCTF), that gradually fuses the tokens based on multi-criteria (e.g., similarity, informativeness, and size of fused tokens). Further, we utilize the one-step-ahead attention, which is the improved approach to capture the informativeness of the tokens. By training the model equipped with MCTF using a token reduction consistency, we achieve the best speed-accuracy trade-off in the image classification (ImageNet1K). Experimental results prove that MCTF consistently surpasses the previous reduction methods with and without training. Specifically, DeiT-T and DeiT-S with MCTF reduce FLOPs by about 44% while improving the performance (+0.5%, and +0.3%) over the base model, respectively. We also demonstrate the applicability of MCTF in various Vision Transformers (e.g., T2T-ViT, LV-ViT), achieving at least 31% speedup without performance degradation. Code is available at https://github.com/mlvlab/MCTF.
Training a linear classifier or lightweight model on top of pretrained vision model outputs, so-called 'frozen features', leads to impressive performance on a number of downstream few-shot tasks. Currently, frozen features are not modified during training. On the other hand, when networks are trained directly on images, data augmentation is a standard recipe that improves performance with no substantial overhead. In this paper, we conduct an extensive pilot study on few-shot image classification that explores applying data augmentations in the frozen feature space, dubbed 'frozen feature augmentation (FroFA)', covering twenty augmentations in total. Our study demonstrates that adopting a deceptively simple pointwise FroFA, such as brightness, can improve few-shot performance consistently across three network architectures, three large pretraining datasets, and eight transfer datasets.
How well the heart is functioning can be quantified through measurements of myocardial deformation via echocardiography. Clinical assessment of cardiac function is generally focused on global indices of relative shortening, however, territorial, and segmental strain indices have shown to be abnormal in regions of myocardial disease, such as scar. In this work, we propose a single framework to predict myocardial disease substrates at global, territorial, and segmental levels using regional myocardial strain traces as input to a convolutional neural network (CNN)-based classification algorithm. An anatomically meaningful representation of the input data from the clinically standard bullseye representation to a multi-channel 2D image is proposed, to formulate the task as an image classification problem, thus enabling the use of state-of-the-art neural network configurations. A Fully Convolutional Network (FCN) is trained to detect and localize myocardial scar from regional left ventricular (LV) strain patterns. Simulated regional strain data from a controlled dataset of virtual patients with varying degrees and locations of myocardial scar is used for training and validation. The proposed method successfully detects and localizes the scars on 98% of the 5490 left ventricle (LV) segments of the 305 patients in the test set using strain traces only. Due to the sparse existence of scar, only 10% of the LV segments in the virtual patient cohort have scar. Taking the imbalance into account, the class balanced accuracy is calculated as 95%. The performance is reported on global, territorial, and segmental levels. The proposed method proves successful on the strain traces of the virtual cohort and offers the potential to solve the regional myocardial scar detection problem on the strain traces of the real patient cohorts.
The task of few-shot image classification and segmentation (FS-CS) involves classifying and segmenting target objects in a query image, given only a few examples of the target classes. We introduce the Vision-Instructed Segmentation and Evaluation (VISE) method that transforms the FS-CS problem into the Visual Question Answering (VQA) problem, utilising Vision-Language Models (VLMs), and addresses it in a training-free manner. By enabling a VLM to interact with off-the-shelf vision models as tools, the proposed method is capable of classifying and segmenting target objects using only image-level labels. Specifically, chain-of-thought prompting and in-context learning guide the VLM to answer multiple-choice questions like a human; vision models such as YOLO and Segment Anything Model (SAM) assist the VLM in completing the task. The modular framework of the proposed method makes it easily extendable. Our approach achieves state-of-the-art performance on the Pascal-5i and COCO-20i datasets.
The integration of artificial intelligence (AI) with radiology marks a transformative era in medicine. Vision foundation models have been adopted to enhance radiologic imaging analysis. However, the distinct complexities of radiologic 2D and 3D radiologic data pose unique challenges that existing models, pre-trained on general non-medical images, fail to address adequately. To bridge this gap and capitalize on the diagnostic precision required in radiologic imaging, we introduce Radiologic Contrastive Language-Image Pre-training (RadCLIP): a cross-modal vision-language foundational model that harnesses Vision Language Pre-training (VLP) framework to improve radiologic image analysis. Building upon Contrastive Language-Image Pre-training (CLIP), RadCLIP incorporates a slice pooling mechanism tailored for volumetric image analysis and is pre-trained using a large and diverse dataset of radiologic image-text pairs. The RadCLIP was pre-trained to effectively align radiologic images with their corresponding text annotations, creating a robust vision backbone for radiologic images. Extensive experiments demonstrate RadCLIP's superior performance in both uni-modal radiologic image classification and cross-modal image-text matching, highlighting its significant promise for improving diagnostic accuracy and efficiency in clinical settings. Our Key contributions include curating a large dataset with diverse radiologic 2D/3D radiologic image-text pairs, a slice pooling adapter using an attention mechanism for integrating 2D images, and comprehensive evaluations of RadCLIP on various radiologic downstream tasks.
In goal-oriented communications, the objective of the receiver is often to apply a Deep-Learning model, rather than reconstructing the original data. In this context, direct learning over compressed data, without any prior decoding, holds promise for enhancing the time-efficient execution of inference models at the receiver. However, conventional entropic-coding methods like Huffman and Arithmetic break data structure, rendering them unsuitable for learning without decoding. In this paper, we propose an alternative approach in which entropic coding is realized with Low-Density Parity Check (LDPC) codes. We hypothesize that Deep Learning models can more effectively exploit the internal code structure of LDPC codes. At the receiver, we leverage a specific class of Recurrent Neural Networks (RNNs), specifically Gated Recurrent Unit (GRU), trained for image classification. Our numerical results indicate that classification based on LDPC-coded bit-planes surpasses Huffman and Arithmetic coding, while necessitating a significantly smaller learning model. This demonstrates the efficiency of classification directly from LDPC-coded data, eliminating the need for any form of decompression, even partial, prior to applying the learning model.
Neural network is a powerful learning paradigm for data feature learning in the era of big data. However, most neural network models are deterministic models that ignore the uncertainty of data. Fuzzy neural networks are proposed to address this problem. FDNN is a hierarchical deep neural network that derives information from both fuzzy and neural representations, the representations are then fused to form representation to be classified. FDNN perform well on uncertain data classification tasks. In this paper, we proposed a novel hierarchical fused quantum fuzzy neural network (HQFNN). Different from classical FDNN, HQFNN uses quantum neural networks to learn fuzzy membership functions in fuzzy neural network. We conducted simulated experiment on two types of datasets (Dirty-MNIST and 15-Scene), the results show that the proposed model can outperform several existing methods. In addition, we demonstrate the robustness of the proposed quantum circuit.
Out-of-context (OOC) detection is a challenging task involving identifying images and texts that are irrelevant to the context in which they are presented. Large vision-language models (LVLMs) are effective at various tasks, including image classification and text generation. However, the extent of their proficiency in multimodal OOC detection tasks is unclear. In this paper, we investigate the ability of LVLMs to detect multimodal OOC and show that these models cannot achieve high accuracy on OOC detection tasks without fine-tuning. However, we demonstrate that fine-tuning LVLMs on multimodal OOC datasets can further improve their OOC detection accuracy. To evaluate the performance of LVLMs on OOC detection tasks, we fine-tune MiniGPT-4 on the NewsCLIPpings dataset, a large dataset of multimodal OOC. Our results show that fine-tuning MiniGPT-4 on the NewsCLIPpings dataset significantly improves the OOC detection accuracy in this dataset. This suggests that fine-tuning can significantly improve the performance of LVLMs on OOC detection tasks.
Medical data often exhibits distribution shifts, which cause test-time performance degradation for deep learning models trained using standard supervised learning pipelines. This challenge is addressed in the field of Domain Generalization (DG) with the sub-field of Single Domain Generalization (SDG) being specifically interesting due to the privacy- or logistics-related issues often associated with medical data. Existing disentanglement-based SDG methods heavily rely on structural information embedded in segmentation masks, however classification labels do not provide such dense information. This work introduces a novel SDG method aimed at medical image classification that leverages channel-wise contrastive disentanglement. It is further enhanced with reconstruction-based style regularization to ensure extraction of distinct style and structure feature representations. We evaluate our method on the complex task of multicenter histopathology image classification, comparing it against state-of-the-art (SOTA) SDG baselines. Results demonstrate that our method surpasses the SOTA by a margin of 1% in average accuracy while also showing more stable performance. This study highlights the importance and challenges of exploring SDG frameworks in the context of the classification task. The code is publicly available at https://github.com/BioMedIA-MBZUAI/ConDiSR